Is a Prestigious Job the same as a Prestigious Country? A Case Study on Multilingual Sentence Embeddings and European Countries
share
We study how multilingual sentence representations capture European countries and how this differs across European languages. We prompt the models with templated sentences that we machine-translate into 12 European languages and analyze the most prominent dimensions in the embeddings. Our analysis reveals that the most prominent country feature in the embedding is its economic strength in terms of GPD. When prompted specifically for job prestige, the embedding space clearly distinguishes high and low-prestige jobs. The occupational dimension is uncorrelated with the most dominant country dimensions for three out of four studied models. One model: Distilled Multilingual Universal Sentence Encoder, however, exhibited a connection between occupational prestige and country of origin, which is a potential source of nationality-based discrimination. Our findings are consistent across languages and, to some extent, with the exception mentioned above, across studied representation models.
READ FULL TEXT